A Balanced Solution for the Partition-based Spatial Merge join in MapReduce

Several MapReduce frameworks have been developed in recent years in order to cope with the need to process an increasing amount of data. Moreover, some extensions of them have been proposed to deal with particular kind of information, like the spatial one. In this paper we will refer to SpatialHadoop, a spatial extension of Apache Hadoop which provides a rich set of spatial data types and operations. In the geo-spatial domain, spatial join is considered a fundamental operation for performing data analysis. However, the join operation is generally classified as a critical task to be performed in MapReduce, since it requires to process two datasets at time. Several different solutions have been proposed in literature for efficiently performing a spatial join which may or may not require the presence of a spatial index computed on both datasets or only one of them. As already discussed in literature, the efficiency of such operation depends on the ability to both prune unnecessary data as soon as possible and to provide a balanced amount of work to be done by each parallelly executed task. In this paper,we take a step forward in this direction by proposing an evolution of the Partition-based Spatial Merge Join algorithm which tries to completely exploit the benefit of the parallelism induced by the MapReduce framework. In particular, we concentrate on the partition phase which has to produce filtered balanced and meaningful subdivisions of the original datasets.